xref: /aosp_15_r20/external/XNNPACK/src/x8-lut/scalar.c.in (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)
1*4bdc9457SAndroid Build Coastguard Worker// Copyright 2021 Google LLC
2*4bdc9457SAndroid Build Coastguard Worker//
3*4bdc9457SAndroid Build Coastguard Worker// This source code is licensed under the BSD-style license found in the
4*4bdc9457SAndroid Build Coastguard Worker// LICENSE file in the root directory of this source tree.
5*4bdc9457SAndroid Build Coastguard Worker
6*4bdc9457SAndroid Build Coastguard Worker$assert BATCH_TILE >= 1
7*4bdc9457SAndroid Build Coastguard Worker$ABC = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"
8*4bdc9457SAndroid Build Coastguard Worker#include <assert.h>
9*4bdc9457SAndroid Build Coastguard Worker
10*4bdc9457SAndroid Build Coastguard Worker#include <xnnpack/lut.h>
11*4bdc9457SAndroid Build Coastguard Worker#include <xnnpack/common.h>
12*4bdc9457SAndroid Build Coastguard Worker
13*4bdc9457SAndroid Build Coastguard Worker
14*4bdc9457SAndroid Build Coastguard Workervoid xnn_x8_lut_ukernel__scalar_x${BATCH_TILE}(
15*4bdc9457SAndroid Build Coastguard Worker    size_t n,
16*4bdc9457SAndroid Build Coastguard Worker    const uint8_t* x,
17*4bdc9457SAndroid Build Coastguard Worker    uint8_t* y,
18*4bdc9457SAndroid Build Coastguard Worker    const uint8_t t[restrict XNN_MIN_ELEMENTS(256)])
19*4bdc9457SAndroid Build Coastguard Worker{
20*4bdc9457SAndroid Build Coastguard Worker  assert(n != 0);
21*4bdc9457SAndroid Build Coastguard Worker  assert(x != NULL);
22*4bdc9457SAndroid Build Coastguard Worker  assert(y != NULL);
23*4bdc9457SAndroid Build Coastguard Worker
24*4bdc9457SAndroid Build Coastguard Worker  $if BATCH_TILE > 1:
25*4bdc9457SAndroid Build Coastguard Worker    for (; n >= ${BATCH_TILE} * sizeof(uint8_t); n -= ${BATCH_TILE} * sizeof(uint8_t)) {
26*4bdc9457SAndroid Build Coastguard Worker      $for N in range(BATCH_TILE):
27*4bdc9457SAndroid Build Coastguard Worker        const size_t vx${N} = (size_t) x[${N}];
28*4bdc9457SAndroid Build Coastguard Worker      x += ${BATCH_TILE};
29*4bdc9457SAndroid Build Coastguard Worker
30*4bdc9457SAndroid Build Coastguard Worker      $for N in range(BATCH_TILE):
31*4bdc9457SAndroid Build Coastguard Worker        const uint32_t vt${N} = (uint32_t) t[vx${N}];
32*4bdc9457SAndroid Build Coastguard Worker
33*4bdc9457SAndroid Build Coastguard Worker      $for N in range(BATCH_TILE):
34*4bdc9457SAndroid Build Coastguard Worker        y[${N}] = (uint8_t) vt${N};
35*4bdc9457SAndroid Build Coastguard Worker      y += ${BATCH_TILE};
36*4bdc9457SAndroid Build Coastguard Worker    }
37*4bdc9457SAndroid Build Coastguard Worker    if XNN_UNLIKELY(n != 0) {
38*4bdc9457SAndroid Build Coastguard Worker      $if BATCH_TILE > 2:
39*4bdc9457SAndroid Build Coastguard Worker        do {
40*4bdc9457SAndroid Build Coastguard Worker          const size_t vx = (size_t) *x++;
41*4bdc9457SAndroid Build Coastguard Worker          const uint32_t vt = (uint32_t) t[vx];
42*4bdc9457SAndroid Build Coastguard Worker          *y++ = (uint8_t) vt;
43*4bdc9457SAndroid Build Coastguard Worker          n -= sizeof(uint8_t);
44*4bdc9457SAndroid Build Coastguard Worker        } while (n != 0);
45*4bdc9457SAndroid Build Coastguard Worker      $else:
46*4bdc9457SAndroid Build Coastguard Worker        const size_t vx = (size_t) *x;
47*4bdc9457SAndroid Build Coastguard Worker        const uint32_t vt = (uint32_t) t[vx];
48*4bdc9457SAndroid Build Coastguard Worker        *y = (uint8_t) vt;
49*4bdc9457SAndroid Build Coastguard Worker    }
50*4bdc9457SAndroid Build Coastguard Worker  $else:
51*4bdc9457SAndroid Build Coastguard Worker    do {
52*4bdc9457SAndroid Build Coastguard Worker      const size_t vx = (size_t) *x++;
53*4bdc9457SAndroid Build Coastguard Worker      const uint32_t vt = (uint32_t) t[vx];
54*4bdc9457SAndroid Build Coastguard Worker      *y++ = (uint8_t) vt;
55*4bdc9457SAndroid Build Coastguard Worker      n -= sizeof(uint8_t);
56*4bdc9457SAndroid Build Coastguard Worker    } while (n != 0);
57*4bdc9457SAndroid Build Coastguard Worker}
58